LM34/LM35
Precision Monolithic
Temperature Sensors
Introduction
Most commonly-used electrical temperature sensors are dif-
ficult to apply. For example, thermocouples have low output
levels and require cold junction compensation. Thermistors
are nonlinear. In addition, the outputs of these sensors are
not linearly proportional to any temperature scale. Early
monolithic sensors, such as the LM3911, LM134 and LM135,
overcame many of these difficulties, but their outputs are
related to the Kelvin temperature scale rather than the more
popular Celsius and Fahrenheit scales. Fortunately, in 1983
two I.C.’s, the LM34 Precision Fahrenheit Temperature Sen-
sor and the LM35 Precision Celsius Temperature Sensor,
were introduced. This application note will discuss the LM34,
but with the proper scaling factors can easily be adapted to
the LM35.
The LM34 has an output of 10 mV/˚F with a typical nonlin-
earity of only
±
0.35˚F over a −50 to +300˚F temperature
range, and is accurate to within
±
0.4˚F typically at room
temperature (77˚F). The LM34’s low output impedance and
linear output characteristic make interfacing with readout or
control circuitry easy. An inherent strength of the LM34 over
other currently available temperature sensors is that it is not
as susceptible to large errors in its output from low level
leakage currents. For instance, many monolithic tempera-
ture sensors have an output of only 1 µA/˚K. This leads to a
1˚K error for only 1 µ-Ampere of leakage current. On the
other hand, the LM34 may be operated as a current mode
device providing 20 µA/˚F of output current. The same 1 µA
of leakage current will cause an error in the LM34’s output of
only 0.05˚F (or 0.03˚K after scaling).
Low cost and high accuracy are maintained by performing
trimming and calibration procedures at the wafer level. The
device may be operated with either single or dual supplies.
With less than 70 µA of current drain, the LM34 has very little
self-heating (less than 0.2˚F in still air), and comes in a
TO-46 metal can package, a SO-8 small outline package
and a TO-92 plastic package.
Forerunners to the LM34
The making of a temperature sensor depends upon exploit-
ing a property of some material which is a changing function
of temperature. Preferably this function will be a linear func-
tion for the temperature range of interest. The base-emitter
voltage (V
BE
) of a silicon NPN transistor has such a tem-
perature dependence over small ranges of temperature.
Unfortunately, the value of V
BE
varies over a production
range and thus the room temperature calibration error is not
specified nor guaranteeable in production. Additionally, the
temperature coefficient of about −2 mV/˚C also has a toler-
ance and spread in production. Furthermore, while the
tempo may appear linear over a narrow temperature, there is
a definite nonlinearity as large as 3˚C or 4˚C over a full
−55˚C to +150˚C temperature range.
Another approach has been developed where the difference
in the base-emitter voltage of two transistors operated at
different current densities is used as a measure of tempera-
ture. It can be shown that when two transistors, Q1 and Q2,
are operated at different emitter current densities, the differ-
ence in their base-emitter voltages, ∆V
BE
,is
(1)
where k is Boltzman’s constant, q is the charge on an
electron, T is absolute temperature in degrees Kelvin and
J
E1
and J
E2
are the emitter current densities of Q1 and Q2
respectively. A circuit realizing this function is shown in Fig-
ure 1.
Equation (1) implies that as long as the ratio of I
E1
to I
E2
is
held constant, then ∆V
BE
is a linear function of temperature
(this is not exactly true over the whole temperature range,
but a correction circuit for the nonlinearity of V
BE1
and V
BE2
will be discussed later). The linearity of this ∆V
BE
with tem-
perature is good enough that most of today’s monolithic
temperature sensors are based upon this principle.
An early monolithic temperature sensor using the above
principle is shown in Figure 2. This sensor outputs a voltage
which is related to the absolute temperature scale by a factor
of 10 mV per degree Kelvin and is known as the LM135. The
circuit has a ∆V
BE
of approximately
(0.2 mV/˚K) x (T)
developed across resistor R. The amplifier acts as a servo to
enforce this condition. The ∆V
BE
appearing across resistor R
is then multiplied by the resistor string consisting of R and
the 26R and 23R resistors for an output voltage of
(10 mV/˚K) x (T). The resistor marked 100R is used for offset
trimming. This circuit has been very popular, but such Kelvin
temperature sensors have the disadvantage of a large con-
stant output voltage of 2.73V which must be subtracted for
use as a Celsius-scaled temperature sensor.
00905101
FIGURE 1.
National Semiconductor
Application Note 460
October 1986
LM34/LM35 Precision Monolithic Temperature Sensors AN-460
© 2002 National Semiconductor Corporation AN009051 www.national.com